# Riemann vs. Stieltjes Integral

From Rudin's PMA:

6.2 Definition Let $\alpha$ be a monotonically increasing function on $[a,b]$. ... Corresponding to each partition $P$ of $[a,b]$, we write

$\Delta \alpha_i = \alpha(x_i) - \alpha(x_{i-1})$.

He then goes on to define the Riemann Stieltjes integral of $f$ with respect to $\alpha$, over the interval $[a,b]$.

The Riemann integral is then pointed out to be a special case of this when $\alpha(x)=x$.

With $\alpha(x)=x$, I understand $\Delta x = x_i - x_{i-1}$ to represent the directed magnitude of the "base of the approximating rectangle" that we then multiply by the value of $f$ taken somewhere within this interval, thus obtaining the area of an approximating rectangle. I don't know where to begin to interpret the case where $\alpha(x) \not\equiv x$.

-
Integrating with respect to a function of bounded variation is a Borel measure. It's a weight function: where a(x) is increasing faster, the more heavily the integral is weighted towards the area under that part of the curve. It's like a CDF, and the integral is weighted wrt to the corresponding density. The idea is that defining it this way gives you a clearer understanding of what things are integrable, because you can think outside the box a bit more; R-S is the natural way to show that all maps from C([a,b])->R are integrals. – Nicholas Wilson Jun 3 '11 at 16:46

The idea is that we define a different way of measuring the width of the rectangle. Instead of using the difference of the two coordinates, we take the difference of the images of those two coordinates under the map $\alpha$. Basically, the idea is that the $(x_i-x_{i-1})$ is to be interpreted as the width or measure of the interval, a construction which leads more generally to the Lebesgue measure of a subset of the reals. The $\alpha(x_i)-\alpha(x_{i-1})$ leads to a different measure, starting with a different assignment of "size" to an interval.

Note that when $\alpha$ is continuously differentiable, then $\alpha(x_i) -\alpha(x_{i-1})$ is equal to $\alpha'(x) (x_i - x_{i-1})$ for $x \in [x_{i-1}, x_i]$ up to higher order terms. So in this case the Stieltjes integral with respect to $\alpha$ is the Riemann integral of the same function times $\alpha'$.

-
Maybe this was inherent in the question, but what are situations where one is more advantageous compared to the other? – Jonathan Fischoff Jul 27 '10 at 4:25
Thanks Akhil. My first exposure to a more abstract measure (different from a distance function) was through Jordan content in the context of 2+ dimensional Riemann integration. Jordan content takes into account topological properties of the set when assigning a measure, specifically - a Jordan region is one where the volume of the boundary is essentially zero (More about this in another question?). So, thank you for invoking the term measure to describe the interpretation of \alpha above, I would have continued to reserve the term for more intimidating things than a monotonic function. – Tom Stephens Jul 27 '10 at 13:00

To add to Akhil's answer, the Stieltjes integral made a lot more sense to me when I started thinking about it in terms of physics. Think of taking something like the integral you get to determine moment of inertia, where it's mass--a function of position--you're integrating over, and not position itself.

-

In Stieltjes integral you assign different importance to different parts of the set you are integrating over.

The usefulness of this would become clearer when you know the theory of Lebesgue measure, which generalizes this even further. For example there is the Dirac measure which when used for integrating functions cares only about the value of a function at a point(typically the origin).

Understanding it in the Stieltjes form is not any harder; moreover it will pave the way for measure theory.

-